Anti-CAR-engineered T cells for epitope-based elimination of autologous CAR T cells

All cell lines were kept at 37 °C in a humidified atmosphere of 5% CO₂. PC3 cells were stably transduced with the open reading frame (orf) encoding the prostate stem cell antigen (PSCA) as previously reported [25]. The murine fibroblast cell line 3T3 and the human embryonic kidney cell line HEK293T were cultured in DMEM (ThermoFisher Scientific, Schwerte, Germany) supplemented with 10% FCS, 100 µg/ml penicillin/streptomycin and 1% non-essential aa. PC3-PSCA and Chinese hamster ovary (CHO) cells were maintained in RPMI 1640 media supplemented with 10% FCS, 100 µg/ml penicillin/streptomycin, 1% non-essential aa, 1 mM sodium pyruvate and 2 mM N-acetyl-l-alanyl-l-glutamine (all purchased from Biochrom, Berlin, Germany).

Cloning and structural composition of conventional CARs as well as the universal chimeric antigen receptor (UniCAR) 28/ζ construct have been described previously in detail [23, 24, 26]. Briefly, they contain an extracellular binding moiety derived from the αPSCA MB1 mAb [25], the αCD19 HD37 mAb [27] or the αLa 5B9 mAb [28] followed by hinge and transmembrane domain of human CD28 and cytoplasmic activation domains of CD28 and CD3ζ. As a targetable moiety, the peptide epitope E7B6 (E-tag) is incorporated in the extracellular spacer region. The Stop variant of the UniCAR lacks intracellular sequences downstream of the CD28 transmembrane domain.

To generate a UniCAR lacking the E-tag, the DNA fragment coding for the extracellular part of the CAR 28/ζ was cut out of the respective vector with SfiI and HpaI. A DNA sequence encoding the scFv 5B9 and hinge region without the E7B6 tag was synthesized by Eurofins Genomics (Ebersberg, Germany) and cloned via SfiI and HpaI into the expression vector p6NST60-CAR 28/ζ, generating the ΔCAR 28/ζ construct.

The novel αE-tag CAR 28/ζ was obtained as follows: first, a fragment encoding the antigen-binding domain was generated by cutting the sequence of the humanized scFv La (7B6) V_L–V_H out of a pSecTag2B vector with NheI and ApaI. The orf of a fragment containing the hinge, transmembrane and signaling domain of human CD28 as well as human CD3ζ signaling chain was cut out of a cloning vector available in our lab using the restriction enzymes MssI and ApaI. Finally, the expression vector p6NST60-MCS was digested with XbaI/KspAI and ligated with the aforementioned two fragments resulting in the vector p6NST60-hu αE-tag CAR 28/ζ.

The Orf of all CAR constructs was C-terminally fused to the DNA sequence coding for enhanced green fluorescent protein (EGFP). To allow for a co-translation of CAR and EGFP from one mRNA, both reading frames are separated by a 2A ‘cleavage’ site derived from Thosea asigna virus that induces a ribosomal ‘skip’ from one codon to the next without the formation of a peptide bond [29].

Design, cloning, and purification of αPSCA-E5B9 TM from 3T3 cell culture supernatants was previously described in detail [25, 30]. For redirection of αE-tag CAR T cells, murine or humanized αPSCA scFvs (clone MB1) were C-terminally modified to contain the E7B6 epitope (E-tag). The respective orfs were cloned into the p6NST50 vector which was used to generate a stable CHO cell line for permanent TM production. Purification from cell culture supernatants and analysis of protein concentration were conducted as reported elsewhere [25, 30, 31].

Buffy coats were supplied by German Red Cross (Dresden, Germany). Human CD3⁺, CD4⁺ or CD8⁺ T cells were isolated and cultured as described previously [25, 28, 30, 31].

Production of lentiviral particles using HEK293T cells and T-cell transduction was conducted as reported elsewhere [23, 26, 32]. In brief, T cells were stimulated using αCD3/CD28 DynaBeads^® (Invitrogen, ThermoFisher Scientific) at a 1:4 bead to cell ratio. For T-cell transduction concentrated virus supernatant was added five times within the first 3 days of expansion. After successful genetic modification verified by EGFP expression, cells were sorted on a FACSAria III (BD Biosciences Pharmingen, Heidelberg, Germany). One day prior to experiments, T cells were rested in complete RPMI 1640 lacking any recombinant cytokines.

Flow cytometry was carried out on a MACSQuant^® Analyzer (Miltenyi Biotec, Bergisch Gladbach, Germany) and acquired data were analyzed using MACSQuantify^® Software (Miltenyi Biotec) or FlowJo 10.1 Software (TreeStar Inc., Ashland, OR USA). Fluorescently labeled mAbs directed against human CD3 (clone BW264/56), CD4 (clone REA623), CD8 (clone REA734), CD69 (clone FN50), granzyme A (GzmA, clone REA162) and granzyme B (GzmB, clone REA226) as well as respective REA isotype controls (clone REA293) were purchased from Miltenyi Biotec. mAbs against human CD107a (clone H4A3) and IgG1 (isotype control, clone MOPC-21) were obtained from BD Biosciences Pharmingen. For the detection of E-tagged CAR constructs on genetically modified T cells, the αE-tag mAb (clone 7B6) and secondary goat anti-mouse IgG F(ab′)2-PE Ab (Beckmann Coulter, Krefeld, Germany) were used [23].

After staining of extracellular markers and live vs. dead cell discrimination applying a Zombie Red™ Fixable Viability Kit (BioLegend, London, UK), T cells were processed using an Inside Stain Kit (Miltenyi Biotec) according to the manufacturer’s protocol and subsequently labeled for intracellular markers.

The elimination of CAR T cells by αE-tag CAR T cells was assessed using a previously established flow cytometry-based assay [33, 34]. To distinguish effector from target cells, the latter were labeled with 10 µM cell proliferation dye eFluor™450 (eBioscience, ThermoFisher Scientific) according to the manufacturer’s instructions. The next day, effector cells (αE-tag CAR T cells) were incubated with eFluor™450-stained CAR 28/ζ, ΔCAR 28/ζ or CAR Stop T cells at indicated effector to target cell (E:T) ratios. After indicated time points, cocultures were carefully resuspended and an aliquot of 25 µl diluted 1:4 with 1 µg/ml propidium iodide (Sigma-Aldrich, Munich, Germany) was measured at a MACSQuant^® Analyzer. After exclusion of doublets and dead cells, cells/ml were assessed for both effector and target T cells separately. Finally, cells of one triplet were pooled, stained for intracellular expression of GzmA and GzmB, and analyzed by flow cytometry.

The antitumor activity of αE-tag CAR T cells was determined in standard chromium-51 release assays as previously reported [25].

A total of 5x10⁵ eFluor™450-labeled CAR 28/ζ, ΔCAR 28/ζ or CAR Stop T cells was cultured together with equal numbers of αE-tag CAR T cells. Additionally, 5 µl of αCD107a mAb or IgG1 isotype control were pipetted to each well. After 1 h of incubation at 37 °C, 1 µl of 2 mM monensin (Sigma-Aldrich) was added. After 20 h, cells were harvested, stained with αCD69 and αCD3 mAb and measured using a MACSQuant^® Analyzer. Living cells were distinguished from dead cells by being propidium iodide negative.

For analysis of TNF, IFN-γ, GM-CSF and IL-2 concentrations in cell-free culture supernatants (stored at − 80 °C) enzyme-linked immunosorbent assay (ELISA) was conducted using OptEIA™ Human ELISA Kits (BD Biosciences Pharmingen) according to the manufacturer’s protocol.

Five-week-old male NMRI-Foxn1^nu/Foxn1^nu mice were s.c. injected into their right flank with 5 × 10⁵ firefly luciferase (Luc)-expressing PC3-PSCA cells (n = 5 per group). Control animals received tumor cells only or tumor cells mixed with either 5 × 10⁵ CAR 28/ζ T cells or 15 × 10⁵ αE-tag CAR T cells. For induction of anti-tumor activity, 10 µg of E5B9- or E7B6-tagged αPSCA TM was administered in addition to tumor cells and CAR-engrafted T cells. To impair the anti-tumor response, one group of mice was transplanted with PC3-PSCA-Luc cells, 10 µg αPSCA-E5B9 TM, 5 × 10⁵ E-tagged CAR target T cells, and 15 × 10⁵ αE-tag CAR effector T cells. Animals were anesthetized 6 h, 1 day and 2 days post-cell mixture injection and bioluminescence imaging was conducted as described elsewhere [26, 32]. Optical images were created with Bruker Multispectral software (Bruker, Germany).

For targeting of CAR-modified T cells, we compared and tested different short peptide tags that can be incorporated into the extracellular part of both our conventional CARs as well as our UniCARs (data not shown). Thereby, we identified a short epitope derived from the nuclear protein La/SS-B (E-tag) as the most suitable tag for our approach. It could be integrated without impairing the in vitro or in vivo functionality of CAR T cells as previously published [23, 24, 26, 27, 32, 35‐38]. The 18-aa-long sequence E7B6 (EKEALKKIIEDQQESLNKW) is specifically bound by the αLa 7B6 mAb [39]. Consequently, we cloned a CAR with a 7B6-derived scFv as the antigen-binding moiety and termed the resulting αCAR construct αE-tag CAR (Fig. 1a). The αE-tag CAR construct recognizes the peptide tag E7B6 located in the hinge region of CARs. Upon antigen recognition, αE-tag CAR effector cells are cross-linked to target cells, which should result in the elimination of the latter. For signal transduction, the newly generated CAR contains the activating cytoplasmic domains of CD3 and CD28. Isolated CD3⁺ T cells from healthy donors could be successfully modified to express the novel CAR with CD4⁺ and CD8⁺ subpopulations yielding comparable transduction rates (supplementary Fig. 1a).

To test functionality of the novel αE-tag CAR construct, flow cytometry-based cytotoxicity assays were performed. As target cells, T lymphocytes were used that express a UniCAR containing the extracellular E-tag and an intracellular 28/ζ signaling domain. As shown in Fig. 1b, culturing these cells together with mock-transduced control cells did not influence viability. In contrast, adding αE-tag CAR effector T cells resulted in significant elimination of target T cells. Expectedly, T-cell killing was dependent on the E:T ratio. To verify that the observed cytotoxic effect is due to specific recognition and binding of the incorporated peptide epitope, experiments with E-tag-deleted CAR T cells (termed ΔCAR 28/ζ) were conducted (Fig. 1c). As anticipated, ΔCAR 28/ζ-engineered lymphocytes were not targeted by αE-tag CAR effector T cells. In addition, we monitored the number of living effector T cells after 24 h and 48 h of coculture. To our surprise, T cells redirected against the E-tag were significantly reduced in cell number whilst viability was maintained in the presence of ΔCAR 28/ζ-armed target cells (Fig. 1d). Interestingly, this effect inversely correlated with the chosen E:T ratio. To confirm these results, we, furthermore, tested whether E-tagged CAR constructs with different antigen specificity can be targeted as well. In accordance with the data obtained for UniCAR-armed target cells, T cells expressing an αCD19 or an αPSCA conventional CAR (containing the extracellular E-tag) were specifically eliminated upon coculture with αE-tag CAR effector T cells already after 24 h (supplementary Fig. 1b). Again, survival of αCAR-redirected effector cells was affected by coculture with target cells in an E:T-dependent manner.

In a next step, we aimed to clarify whether the diminished effector T-cell number was merely a result of T-cell activation/exhaustion or reflects apoptosis actively induced by cross-linked target cells. For that purpose, expression of activation-induced CD69 as well as CD107a was evaluated (Fig. 2). The latter not only represents a marker for degranulation but was also recently described as a marker protein for fratricide of T cells [40]. In accordance with cytotoxicity data, αE-tag CAR effector T cells selectively upregulated both CD69 and CD107a in the presence of E-tag-labeled CAR 28/ζ target cells but not upon coculture with ΔCAR 28/ζ-equipped lymphocytes (Fig. 2b). Regarding CD69 and CD107a on the target cell side, T cells lacking the peptide epitope did not increase expression of neither of the analyzed surface molecules (Fig. 2c). However, target cells harboring the E-tag were stimulated and degranulated upon coincubation with αE-tag CAR effector T cells. These results indicate that cross-linkage of effector and target T cells triggers activation of both cell types and subsequent reciprocal T-cell elimination.

The data obtained so far clearly demonstrate functionality of the novel αE-tag CAR construct. However, binding to E-tag-comprising CAR 28/ζ target cells led to fratricide especially at low E:T ratios. This could either be due to cross-linkage of both cell types and formation of an immune synapse-like structure per se or due to triggering of the intracellular 28/ζ signaling domain of CAR-equipped target cells. To shed light on this issue, we performed cytotoxicity assays with target T cells that express an E-tag-containing CAR construct without an intracellular signaling unit (CAR Stop) (Fig. 3). As depicted in Fig. 3a, these cells were efficiently depleted by αE-tag CAR effector T cells at all E:T ratios tested, thereby verifying the utility of the E-tag for specific targeting of genetically modified cells. As effector cell numbers remained constant, no reciprocal killing occurred (Fig. 3b). In line with this observation, only redirected effector T cells but not E-tag-labeled target cells were activated and degranulated upon cross-linkage (Fig. 3c). Hence, these data clearly demonstrate that reciprocal elimination of CAR-armed T cells occurs in dependence on an intracellular signaling domain.

To investigate which cell type prevails in coculture, long-term cytotoxicity assays were conducted (supplementary Fig. 2a and 2b). The outcome was both dependent on the targeted CAR construct as well as the E:T ratio. Cells modified to express an E-tag-containing αCD19 or αPSCA conventional CAR were efficiently eradicated long term at E:T ratios of 3:1 and 1:1. Compared to this, UniCAR target cells were completely eliminated only at the highest E:T ratio tested. Thus, the prerequisite for an efficient T-cell-based αCAR therapy is that the E:T ratio is in favor of the αE-tag effector T cells.

In previous studies, we observed that the expression of co-translated EGFP marker protein directly correlates with surface expression of CAR constructs [e.g., 23, 24]. Usually, we verify CAR T-cell expression via an αE-tag mAb (supplementary Fig. 3). As αE-tag CARs lack any extracellular tags, we monitored EGFP signal of T cells as a surrogate for CAR expression level changes (Fig. 4a). Upon activation, effector cells strongly upregulated the expression of their αE-tag CAR as shown for cocultivation with CAR Stop-modified target cells at all tested E:T ratios as well as for a 3:1 ratio with CAR 28/ζ-armed T cells. Vice versa, target cells that receive an activating signal via their intracellular 28/ζ domain upon cross-linkage to effector cells also increased CAR expression (Fig. 4a, lower panel, 1:1 and 1:3 ratio). When analyzing the remaining E-tag-labeled T cells it becomes obvious that they showed reduced EGFP levels compared to conditions in which they were cultured in the absence of effector cells [e.g., CAR Stop: 9.6 vs. 1.0, 2.0 and 2.5 median fluorescence intensity (MFI) of EGFP at 3:1, 1:1 and 1:3 ratio, respectively]. This might be caused by the downregulation of the E-tag-containing CAR, yet, a more likely explanation is the preferential killing of T cells displaying a high CAR expression than of cells expressing only low levels of E-tag-comprising CARs. To substantiate this hypothesis, we analyzed elimination kinetics of target cells that were presorted into CAR low- and CAR high-expressing subpopulations based on the strength of their EGFP signal (supplementary Fig. 4). Compared to their EGFP^low counterpart, target T cells exhibiting high CAR densities were indeed more efficiently killed. In accordance, expansion of αE-tag CAR effector T cells that were incubated with EGFP^high target T cells was more pronounced than in cocultures with sorted EGFP^low cells. Together, these data support our assumption that high E-tagged CAR expression results in preferential target cell killing.

As we established a functional αCAR CAR for efficient depletion of genetically modified T cells, we wanted to use this construct as a tool for investigating CD4⁺ and CD8⁺ T-cell killing capabilities in more detail. For that purpose, we isolated and transduced both subpopulations separately with either an E7B6-tagged CAR Stop or an αE-tag CAR. In subsequent flow cytometry-based cytotoxicity assays over 48 h, both CD4⁺ and CD8⁺ effector cells performed equally well in killing of E-tag-containing CAR target T cells (Fig. 4b). However, the CD4⁺ T-cell subset demonstrated a stronger proliferative capacity compared to CD8⁺ T cells, most likely due to the fact that the latter produce less growth-promoting IL-2 than T-helper cells. Again, monitoring the EGFP signal indicates that effector cells vigorously enhanced the expression of their αE-tag CAR upon activation (supplementary Fig. 5a) and preferentially eliminated target cells with the high expression of the E-tag containing CAR (supplementary Fig. 5b). The cytotoxicity data further reveal that CD4⁺ and CD8⁺ cells can be targeted efficiently with no apparent differences between the two T-cell populations (Fig. 4b).

To shed some light on the mode of action, intracellular staining for granzymes were conducted (supplementary Fig. 6a). As CD8⁺ T cells were already highly positive for GzmA in a resting state, upon cocultivation with target cells only GzmB was strongly upregulated. In contrast, CD4⁺ T-cell activation induced expression of both GzmA and GzmB. Interestingly, we could detect elevated granzyme levels in target cells (supplementary Fig. 6b). As they were not expressing CD107a (see also Fig. 3c), we reason that GzmA and GzmB are not produced by target cells themselves but originated from effector cells and were transported via the cytolytic immune synapse. Further evidence to support this notion was obtained by staining target T cells that were cocultured with mock-transduced T cells instead of cytolytic effector cells. As shown in supplementary Fig. 6c, enhanced GzmA/GzmB was only detectable in E-tagged CAR T cells that were targeted by αE-tag CAR effector T cells, whereas granzyme levels of target T cells cocultured with mock control resembled baseline expression.

Having successfully shown that αCAR-engineered T cells attack E-tagged CAR T cells, we next wanted to clarify how this influences effector functions of the targeted lymphocytes. As readout, secretion of proinflammatory cytokines in vitro as well as tumor cell killing in vivo was chosen. As already stated above, one kind of E-tagged CARs are our switchable and flexible modular UniCARs [24, 26, 27, 32, 35‐38]. Utilizing the La/SS-B-derived peptide epitope E5B9 as linker, the platform technology enables a precisely controlled activation of otherwise inert 5B9 UniCAR-armed T cells by infusion of small E5B9-tagged TMs (Fig. 5a, left panel). For proof-of-concept, we used these UniCAR cells as targets for αCAR-equipped cells and stimulated them against the tumor-associated antigen PSCA via an αPSCA-E5B9 TM. Activation resulted in the substantial release of proinflammatory TNF, GM-CSF and IFN-γ as well as mitogenic IL-2 (Fig. 5b, striped bar). Addition of αE-tag CAR T cells resulted in a dose-dependent reduction of secreted cytokines. Interestingly, retargeting of αCAR T cells against the E-tag did not trigger profound cytokine production (Fig. 5b and supplementary Fig. 7) which, in light of a putative clinical translation, is beneficial since it can be reasoned that T-cell-mediated CAR T-cell killing does not exacerbate cytokine release syndrome. As depicted in Fig. 5c, E-tagged UniCAR 28/ζ T cells were not only eliminated in vitro but also in vivo. NMRI^nu/nu mice were transplanted with Luc-expressing PC3-PSCA cells and survival of tumor cells was monitored by means of bioluminescence imaging. Two days after the injection of Luc⁺ PC3-PSCA cells, UniCAR 28/ζ T cells and a cross-linking αPSCA-E5B9 TM, Luc activity was almost completely cleared. By contrast, anti-tumor activity was markedly impaired upon co-injection of αE-tag CAR T cells (Fig. 5c, right panel) verifying that αCAR-modified T cells are functional in experimental mice post-transfer and efficiently kill E-tagged target T cells in vivo.

The complete eradication of a therapeutic CAR T-cell population inevitably entails termination of antitumor activity. In contrast to other currently available safety strategies, the αCAR CAR methodology provides the unique possibility to re-initiate tumor treatment as the αE-tag CAR T cells can be repurposed as tumor-fighting CAR T cells. As schematically summarized in Fig. 5a, they may even replace the original therapeutic cell product and function as an alternative UniCAR variant which we termed ‘7B6 UniCAR’ T cells. Like the original UniCARs which are directed against the peptide epitope E5B9, 7B6 UniCARs can be redirected by administering E-tag-comprising TMs (Fig. 5a, right panel). To support this idea, we generated two novel TMs consisting of either a murine or a humanized scFv against PSCA fused to the E-tag. Indeed, as demonstrated in Fig. 6a, these TMs efficiently mediated cross-linkage of αE-tag CAR T cells and PSCA-expressing PC3 cells, resulting in significant tumor cell lysis in vitro. The TM-dependent anti-tumor response of αE-tag CAR-modified T cells could furthermore be substantiated in vivo (Fig. 6b). To that end, athymic nude mice were injected with Luc-expressing PSCA⁺ tumor cells either alone or together with αE-tag CAR T cells with or without the humanized PSCA-binding E-tagged TM. After 2 days, Luc activity was no longer detectable in the treatment group (Fig. 6b, right panel) whilst tumor cells were still visible in both control groups (Fig. 6b, left and middle panel). Taken together, αE-tag CAR T cells can not only trigger elimination of autologous CAR T cells but can function as an alternative UniCAR variant if required.